BEHAV3D Tumor Profiler - Guided Tutorial

Table of contents

• Introduction
• What data can I use?
• Importing data
• Joint data processing
  • Scaling
  • PCA analysis
• Modules
  • Heterogeneity Module
  • Large-scale phenotyping Module
  • Small-scale phenotyping Module

Introduction

Welcome to the BEHAV3D Tumor Profiler Demo. Here you can find a demo example of how the Behaved Tumor Profiler pipeline works and what each of the outputs can provide.
Tu run the pipeline for yourself, please refer to the Github page, where you can find a wider explanation of the pipeline, as well as the Google Colab and R script versions.

Firstly all the necessary libraries must be installed

Which libraries do I need?

library(dplyr)
library(dtwclust)
library(e1071)
library(emmeans)
library(ggplot2)
library(gtools)
library(parallel)
library(pheatmap)
library(plotly)
library(plyr)
library(RANN)
library(readr)
library(reshape2)
library(scales)
library(sp)
library(spatstat)
library(stats)
library(tidyr)
library(umap)
library(viridis)
library(zoo)

We set a seed for reproducibility:

set.seed(123)

❔ What data can I use?

Input data can be obtained from image analysis software like Imaris (Oxford Instruments), where you extract the relevant features for each cell types. Each cell type must be inputted separately into the pipeline.

Input data must follow the following format: Mouse_CellType_timelapse_LargeScaleRegion_Feature.csv

Our analyzed Cell types
Tumor Cells	SR101	MG (Microglia)	BV (Blood Vessel)

Example: 2430F13_SR101_timelapse_CL3_2_Distance.csv

Mouse information includes: 2430 -> Mother | F -> Sex | 13 -> Day

LargeScaleRegion information includes: CL2 -> Class (Environmental Cluster) | 2430F13.CL2_2 or just CL2_2 -> Position

These are the features to be uploaded:

• Displacement
• Distance_tumor
• Displacement_Delta_Length
• Displacement_Length
• Position
• Speed
• Time

How does input data look like?

The data is outputed by softwares like Imaris in a .csv format.

For example, a Tumor cells 2430F11 CL1_1 displacement file would look like this (Each row represents a single cell track):

And a SR101 2430F16 CL1_1 Position file would look like this:

📂 Importing data

There are two ways to start or continue your analysis in BEHAV3D Tumor Profiler:

1. You can import your data and start your analysis from the beginning

Data import

We need to import the dataset and classifying the information into distinct dataframes
Create a reading function, that reads input from Imaris software:

## function to import the stats of the tracked tumor cells data from csv files extracted by Imaris
  read_plus <- function(flnm) {
    read_csv(flnm, skip = 3, col_types = cols()) %>% 
      mutate(filename = flnm)
  }

Now we can import the datafiles of interest. We are basically extracting the information from .csv files and creating the corresponding dataframes

Tumor cells information

## set directory where csv files are located
  library(plyr)
  working_directory <- "D:/BEHAV3D_Tumor_Profiler/data/Tracked_tumor_cells"
  setwd(working_directory)
  # import volume per organoid object
  pat = "*Displacement"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  files<-files[seq(1, length(files), 3)]
  disp_csv <- ldply(files, read_plus)
  # import distance to tumor
  pat = "*Distance_tumor"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  dist_tumor_csv <- ldply(files, read_plus)
  # import area per organoid object
  pat = "*Displacement_Delta_Length"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  dipl_delta_csv <- ldply(files, read_plus)
  
  # import area per organoid object
  pat = "*Displacement_Length"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  dipl_length_csv <- ldply(files, read_plus)
  # import position per organoid object
  pat = "*Position"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  pos_csv <- ldply(files, read_plus)
  
  # import speed
  pat = "*Speed"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  speed_csv <- ldply(files, read_plus)
  
  # import Time
  pat = "*Time"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  time_csv <- ldply(files, read_plus)
  
  ### ## make each ID unique (IDs imported from different imaging files can be repeated):
  category <- as.factor(dist_tumor_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  dist_tumor_csv <- left_join(dist_tumor_csv, ranks) 
  dist_tumor_csv$ID2 <- with(dist_tumor_csv, interaction(ID, ranks))
  
  category <- as.factor(pos_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  pos_csv <- left_join(pos_csv, ranks) 
  pos_csv$ID2 <- with(pos_csv, interaction(ID, ranks))
  
  category <- as.factor(time_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  time_csv <- left_join(time_csv, ranks) 
  time_csv$ID2 <- with(time_csv, interaction(ID, ranks))
  
  category <- as.factor(speed_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  speed_csv <- left_join(speed_csv, ranks) 
  speed_csv$ID2 <- with(speed_csv, interaction(ID, ranks))
  
  category <- as.factor(dipl_length_csv$filename)  ##this measures how much distnace did a cell move over, a cell can mover over a big distance and stay at the same position
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  dipl_length_csv <- left_join(dipl_length_csv, ranks) 
  dipl_length_csv$ID2 <- with(dipl_length_csv, interaction(ID, ranks))
  
  category <- as.factor(dipl_delta_csv$filename)  ## this variable measures how much distance did the cell move from its origin
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  dipl_delta_csv <- left_join(dipl_delta_csv, ranks) 
  dipl_delta_csv$ID2 <- with(dipl_delta_csv, interaction(ID, ranks))
  
  category <- as.factor(disp_csv$filename)  ## a bit similar to the previous ones, this measures the area that a cell covers overtime
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  disp_csv <- left_join(disp_csv, ranks) 
  disp_csv$ID2 <- with(disp_csv, interaction(ID, ranks))

Blood vessel (BV) information

Firstly, we create the dataframe containing the statistics of interest 

detach(package:plyr)
  
  ##create dataframe containing the statistics of interest
  master_1 <- left_join(dist_tumor_csv[,c(1,13)],time_csv[,c(1,5,6,8,10,11)], 
                        by=c("ID2" = "ID2"))
  master_1 <- left_join(master_1,disp_csv[,c(1,11)], 
                        by=c("ID2" = "ID2"))
  master_1 <- left_join(master_1,dipl_delta_csv[,c(1,11)], 
                        by=c("ID2" = "ID2"))
  master_1 <- left_join(master_1,dipl_length_csv[,c(1,11)], 
                        by=c("ID2" = "ID2"))
  master_1 <- left_join(master_1,speed_csv[,c(1,11)], 
                        by=c("ID2" = "ID2"))
  master <- left_join(master_1,pos_csv[,c(1,2,3,14)], 
                      by=c("ID2" = "ID2"))
  
  colnames(master) <- c("dist_tumor","ID2","Time","Tracked","TrackID", "filename","ranks","disp2","disp_d","disp_l","speed","x","y","z")

  # Extract mouse data (ID) using regexp (mother+sex+day)
  master$mouse <- master$filename
  master$mouse <- gsub("^.*[_/](\\d*[MF]\\d+).*", "\\1", master$mouse, perl=TRUE)
  
  # Extract position metadata
  master$position <- master$filename
  master$position <- gsub(".*_(CL\\d+_\\d+)_.*", "\\1", master$position, perl=TRUE)
  master$position<-with(master, interaction(mouse, position))
  
  
  # Extract class metadata
  master$class <- master$position
  master$class <- gsub(".*(CL\\d+)_.*", "\\1", master$class, perl=TRUE)
  
  master$filename<-NULL
  ## Round the time
  master$Time<-round(master$Time, digits=2)
  
  
  ##check that the imported tracks look ok
  g<-ggplot(master)+geom_path(aes(x, y, group=TrackID), size=0.2,
       arrow = arrow(ends = "last",type = "open",length = unit(0.01, "inches")))+
      theme_bw()+theme(aspect.ratio = 1)+facet_wrap(class~position,scales = "free")
  
  g

, the Blood Vessel information is extracted:

###Import BLOOD VESSELS stats (does not exist for all tracked positions so needs to be processed separtely.)
  working_directory2 <- "D:/BEHAV3D_Tumor_Profiler/data/BV_stats"
  setwd(working_directory2)
  
  files <- list.files(path = working_directory2, full.names = T, recursive = TRUE)
  BV_df <- ldply(files, read_plus)
  
  ###Process blood vessels data:
  BV_df<-BV_df[c(1,6,7,8,9,11)]
  colnames(BV_df) <- c("dist_BV","Time","Tracked","TrackID", "ID","filename")
  
  ## rename mouse accroding to filename:
  BV_df$mouse<-BV_df$filename
  BV_df$mouse <- gsub("^.*[_/](\\d*[MF]\\d+).*", "\\1", BV_df$mouse, perl=TRUE)
  
  BV_df$position <- BV_df$filename
  BV_df$position <- gsub(".*_(CL\\d+_\\d+)_.*", "\\1", BV_df$position, perl=TRUE)
  BV_df$position <- with(BV_df, interaction(mouse, position))
  
  BV_df$class <- BV_df$position
  BV_df$class <- gsub(".*(CL\\d+)_.*", "\\1", BV_df$class, perl=TRUE)
  
  
  ## create ranks column based on master ranks:
  ranks_master<-subset(master, position %in% BV_df$position)
  ranks_master<-ranks_master[!duplicated(ranks_master$position),c(6,15)]
  colnames(ranks_master)[1]<-"ranks_bv"
  BV_df<- left_join(BV_df,ranks_master)
  
  ## keep only tracked cells and with a TrackID:
  BV_df<-subset(BV_df, Tracked=="Tracked")
  ## remove any NA:
  BV_df<-na.omit(BV_df)
  
  ## create unique Track_ID for Blood vessels dataframe:
  BV_df$Track2 <- with(BV_df, interaction(ranks_bv, TrackID))
  BV_df$Track2 <- gsub(".", '', BV_df$Track2, fixed = T)
  BV_df$Track2 <- as.numeric(as.character(BV_df$Track2))
  ##sumarize per TrackID the average, min and max distance to blood vessels
  ggplot(BV_df,aes(dist_BV))+geom_histogram()+facet_wrap(.~ position)
  BV_df<-na.omit(BV_df)
  BV_df$BV_contact<-ifelse(BV_df$dist_BV<4, 0, 1)
  BV_df_sum<-BV_df%>%group_by(Track2)%>%summarise(BV_sd=sd(dist_BV)/sqrt(length(dist_BV)),BV_mean=mean(dist_BV), BV_min=min(dist_BV), BV_max=max(dist_BV), BV_contact=mean(BV_contact))

SR101 information

library(plyr)

##import Sr101 cells for the populations that have them
  
  working_directory <- "D:/BEHAV3D_Tumor_Profiler/data/SR101_objects"
  setwd(working_directory)
  # import volume per organoid object
  
  # import position per organoid object
  pat = "*Position"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  pos_csv <- ldply(files, read_plus)
  
  
  # import Time
  pat = "*Time"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  time_csv <- ldply(files, read_plus)
  
  ### ## make each TrackID unique (TrackIDs imported from different imaging files can be repeated):
  
  category <- as.factor(pos_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  pos_csv <- left_join(pos_csv, ranks) 
  pos_csv$ID2 <- with(pos_csv, interaction(ID, ranks))
  
  category <- as.factor(time_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  time_csv <- left_join(time_csv, ranks) 
  time_csv$ID2 <- with(time_csv, interaction(ID, ranks))
  
  detach(package:plyr)
  
  ##create dataframe containing the statistics of interest
  master_SR101 <- left_join(pos_csv[,c(1,2,3,10,12)],time_csv[,c(1,9)],
                            by=c("ID2" = "ID2"))
  
  
  colnames(master_SR101) <- c("x","y","z","filename","ID2","Time")
  
  ## rename mouse accroding to filename:
  # Extract mouse data (ID) using regexp (mother+sex+day)
  master_SR101$mouse <- master_SR101$filename
  master_SR101$mouse <- gsub("^.*[_/](\\d*[MF]\\d+).*", "\\1", master_SR101$mouse, perl=TRUE)
  
  # Extract position metadata
  master_SR101$position <- master_SR101$filename
  master_SR101$position <- gsub(".*_(CL\\d+_\\d+)_.*", "\\1", master_SR101$position, perl=TRUE)
  master_SR101$position<-with(master_SR101, interaction(mouse, position))
  
  # Extract class metadata
  master_SR101$class <- master_SR101$position
  master_SR101$class <- gsub(".*(CL\\d+)_.*", "\\1", master_SR101$class, perl=TRUE)
  
  master_SR101$filename<-NULL
  
  ## Convert imported time to hours:
  master_SR101$Time<-round(master_SR101$Time, digits=2)

Microglia (MG) information

##import MG cells for the populations that have them
  library(plyr)
  working_directory <- "D:/BEHAV3D_Tumor_Profiler/data/MG_objects"
  setwd(working_directory)

  # import position per organoid object
  pat = "*Position"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  pos_csv <- ldply(files, read_plus)
  
  
  # import Time
  pat = "*Time"
  files <- list.files(path = working_directory, pattern = pat, full.names = T, recursive = TRUE)
  time_csv <- ldply(files, read_plus)
  
  ### ## make each TrackID unique (TrackIDs imported from different imaging files can be repeated):
  
  category <- as.factor(pos_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  pos_csv <- left_join(pos_csv, ranks) 
  pos_csv$ID2 <- with(pos_csv, interaction(ID, ranks))
  
  category <- as.factor(time_csv$filename)
  ranks <- rank(-table(category), ties.method="first")
  ranks <- as.data.frame(ranks)
  ranks$filename <- row.names(ranks)
  time_csv <- left_join(time_csv, ranks) 
  time_csv$ID2 <- with(time_csv, interaction(ID, ranks))
  
  detach(package:plyr)
  
  ##create dataframe containing the statistics of interest
  master_MG <- left_join(pos_csv[,c(1,2,3,10,12)],time_csv[,c(1,9)], 
                        by=c("ID2" = "ID2"))
  
  
  colnames(master_MG) <- c("x","y","z","filename","ID2","Time")
  
  ## rename mouse accroding to filename:
  # Extract mouse data (ID) using regexp (mother+sex+day)
  
  master_MG$mouse<-master_MG$filename
  master_MG$mouse <- gsub("^.*[_/](\\d*[MF]\\d+).*", "\\1", master_MG$mouse, perl=TRUE)
  
  # Extract position metadata
  master_MG$position <- master_MG$filename
  master_MG$position <- gsub(".*_(CL\\d+_\\d+)_.*", "\\1", master_MG$position, perl=TRUE)
  master_MG$position<-with(master_MG, interaction(mouse, position))
  
  # Extract class metadata
  master_MG$class <- master_MG$position
  master_MG$class <- gsub(".*(CL\\d+)_.*", "\\1", master_MG$class, perl=TRUE)
  
  master_MG$filename<-NULL
  
  ## Convert imported time to hours:
  master_MG$Time<-round(master_MG$Time, digits=2)

Pre-processing

Now we need to join the information from all of the sources into a single dataframe

Interactions between cells

We first calulate the interactions between the nearest cells:

### test that tracks imported are fine:
  master_test<-master%>%group_by(position)%>%mutate(x=x-min(x), y=y-min(y))
  g<-ggplot(master)+geom_path(aes(x, y, group=TrackID, colour=dist_tumor), arrow = arrow(ends = "last", type = "closed",length = unit(0.005, "inches")))+theme_bw()
  g<-g + theme(legend.position = "none")+facet_wrap(.~position, scales = "free")
  g

  ## To calculate interactions between cells 

  List2=list()
  for (m in unique(master$position)){
    distance_M4<-master[which(master$position==m),]
    
    List = list()
    for (i in unique(distance_M4$Time)){
      distancei <- distance_M4[distance_M4$Time==i,]
      distancei2<-as.data.frame(distancei[,c(3,11,12,13)])
      coordin<-ppx(distancei2, coord.type=c("t","s","s", "s"))
      dist1<- nndist(coordin)
      dist2<- nndist(coordin, k=2)
      dist3<- nndist(coordin, k=3)
      dist4<- nndist(coordin, k=4)
      dist5<- nndist(coordin, k=5)
      dist6<- nndist(coordin, k=6)
      dist7<- nndist(coordin, k=7)
      dist8<- nndist(coordin, k=8)
      dist9<- nndist(coordin, k=9)
      dist10<- nndist(coordin, k=10)
      dist<-cbind(dist1,dist2,dist3, dist4,dist5, dist6, dist7, dist8, dist9, dist10)  ### calculate distance to the three nearest neighbors
      dist<-data.frame(rowMeans(dist[,c(1:3)]),rowMeans(dist))
      distanceDFi_dist<-cbind(distancei, dist)
      List[[length(List)+1]] <-distanceDFi_dist ## store to list object
    }
    master_dist <- do.call(rbind, List) ## convert List to dta
    List2[[length(List2)+1]] <-master_dist
  }
  master_distance<-do.call(rbind, List2)
  colnames(master_distance)[c(17,18)] <- c("dist_3_neigh", "dist_10_neigh")
  
  ###remove the untracked tracks:
  master_distance<-subset(master_distance, Tracked=="Tracked")
  
  ## create a unique track_ID for master:
  master_distance$Track2 <- with(master_distance, interaction(ranks, TrackID))
  master_distance$Track2 <- gsub(".", '', master_distance$Track2, fixed = T)
  master_distance$Track2 <- as.numeric(as.character(master_distance$Track2))
  
  
  ##check that the imported tracks look ok
  g<-ggplot(master_distance)+geom_path(aes(x, y, group=Track2), size=0.2,
                              arrow = arrow(ends = "last",type = "open",length = unit(0.01, "inches")))+
    theme_bw()+theme(aspect.ratio = 1)+facet_wrap(class~position,scales = "free")
  
  g

For the SR101 and MG datasets, we separately find the distance to tumor cells

For SR101

##For SR101
  master_distance_SR101<-subset(master_distance, mouse %in% unique(master_SR101$mouse))
  ### Calculate distance from Tumor cells to SR101.
  ### Considering that olig2 doens;t move much, let's project the position of the birghtest timepoint from each movie to all timepoints.
  ### Identify birghtest timepoint by the number of SR101 objects:
  brightest_SR101<-master_SR101%>%group_by(Time, position)%>%summarise(n=n())%>%group_by( position)%>%filter(n==max(n))
  master_SR101<-subset(master_SR101, Time %in%brightest_SR101$Time)
  
  List2<-list()
  for (m in unique(master_distance_SR101$position)){
    master_pos <-master_distance_SR101%>%filter(position==m)%>%group_by(Time)%>%arrange(Time)
    master_SR101_pos <-master_SR101%>%filter(position==m)%>%group_by(Time)%>%arrange(Time)
    List1 = list() ## create list for ID of the contacting objects
    for(t in unique(master_pos$Time)){
      master_pos_t <-master_pos%>%filter(Time==t)
      ### calculate distance to 5 neigrest neighorbing SR101 in a radius of 40um
      cells_radius<-nn2(data=master_SR101_pos[c("x","y","z")], query = master_pos_t[c("x","y","z")],k=30, treetype = 'bd',searchtype =  "radius", radius = 30)
      cells_min<-nn2(data=master_SR101_pos[c("x","y","z")], query = master_pos_t[c("x","y","z")],k=1, treetype = 'bd',searchtype =  "standard")
      
      dist_table = data.frame(master_pos_t[c("Time","Track2","mouse","position","class")],cells_min[["nn.dists"]],cells_radius[["nn.dists"]])
      List1[[length(List1)+1]] <-dist_table  
      rm(temp)
    }
    cell_radius_t <- do.call(rbind, List1)
    List2[[length(List2)+1]] <-cell_radius_t 
    rm(cell_radius_t)
  }
  
  master_dist_SR101 <- do.call(rbind, List2)
  ### convert the distant cells to 0
  
  temp2<-master_dist_SR101[c("X1","X2","X3","X4","X5", "X6", "X7", "X8", "X9", "X10", "X11","X12","X13","X14","X15","X16","X17","X18","X19","X20", "X21","X22","X23","X24","X25","X26","X27","X28","X29","X30")]
  temp2[temp2 >1.340781e+153]<-0
  
  ### Calculate number of contacts:
  master_dist_SR101$n_SR101<-rowSums(temp2 != 0)
  ### renamen min_SR101:
  colnames(master_dist_SR101)[colnames(master_dist_SR101) == "cells_min...nn.dists..."] <- "min_SR101"
  
  ### bind information on SR101 distnaces to the main dataframe:
  master_dist_SR101[c("X1","X2","X3","X4","X5", "X6", "X7", "X8", "X9", "X10", "X11","X12","X13","X14","X15","X16","X17","X18","X19","X20", "X21","X22","X23","X24","X25","X26","X27","X28","X29","X30")]<-NULL
  ## For each Track2 find what is the minimal distance to SR101 and the max n of SR101 neiighbors in 30um diameter:
  master_distance_SR101<-master_dist_SR101%>%group_by(Track2, mouse, position, class)%>%summarize(n_SR101=max(n_SR101), min_SR101 =min(min_SR101))
  ### plot per position types the values of 
  p <- ggplot(master_distance_SR101)+geom_jitter(aes(x=class, y=n_SR101))+
    geom_bar(aes(x=class , y=n_SR101),alpha=0.5,stat = "summary") +ggtitle("number of neighboring SR101 cells per position type")+theme_classic()+
    theme(aspect.ratio=0.7)
  
  p

  p <- ggplot(master_distance_SR101)+geom_jitter(aes(x=class, y=min_SR101))+
    geom_boxplot(aes(x=class , y=min_SR101),alpha=0.5, outlier.colour = NA) +ggtitle("min distance to SR101 cells per position type")+theme_classic()+
    theme(aspect.ratio=0.7)
  
  p

For MG

##For MG
  master_distance_MG<-subset(master_distance, mouse %in% unique(master_MG$mouse))
  ### Calculate distance from Tumor cells to MG.
  ### Considering that olig2 doens;t move much, let's project the position of the birghtest timepoint from each movie to all timepoints.
  ### Identify birghtest timepoint by the number of MG objects:
  brightest_MG<-master_MG%>%group_by(Time, position)%>%summarise(n=n())%>%group_by( position)%>%filter(n==max(n))
  master_MG<-subset(master_MG, Time %in%brightest_MG$Time)
  
  List2<-list()
  for (m in unique(master_distance_MG$position)){
    master_pos <-master_distance_MG%>%filter(position==m)%>%group_by(Time)%>%arrange(Time)
    master_MG_pos <-master_MG%>%filter(position==m)%>%group_by(Time)%>%arrange(Time)
    List1 = list() ## create list for ID of the contacting objects
    for(t in unique(master_pos$Time)){
      master_pos_t <-master_pos%>%filter(Time==t)
      ### calculate distance to 5 neigrest neighorbing MG in a radius of 30um
      cells_radius<-nn2(data=master_MG_pos[c("x","y","z")], query = master_pos_t[c("x","y","z")],k=30, treetype = 'bd',searchtype =  "radius", radius = 30)
      cells_min<-nn2(data=master_MG_pos[c("x","y","z")], query = master_pos_t[c("x","y","z")],k=1, treetype = 'bd',searchtype =  "standard")
      
      dist_table = data.frame(master_pos_t[c("Time","Track2","mouse","position","class")],cells_min[["nn.dists"]],cells_radius[["nn.dists"]])
      List1[[length(List1)+1]] <-dist_table  
      rm(temp)
    }
    cell_radius_t <- do.call(rbind, List1)
    List2[[length(List2)+1]] <-cell_radius_t 
    rm(cell_radius_t)
  }
  
  master_dist_MG <- do.call(rbind, List2)
  ### convert the distant cells to 0
  
  temp2<-master_dist_MG[c("X1","X2","X3","X4","X5", "X6", "X7", "X8", "X9", "X10", "X11","X12","X13","X14","X15","X16","X17","X18","X19","X20", "X21","X22","X23","X24","X25","X26","X27","X28","X29","X30")]
  temp2[temp2 >1.340781e+153]<-0
  
  ### Calculate number of contacts:
  master_dist_MG$n_MG<-rowSums(temp2 != 0)
  ### renamen min_MG:
  colnames(master_dist_MG)[colnames(master_dist_MG) == "cells_min...nn.dists..."] <- "min_MG"
  
  ### bind information on MG distnaces to the main dataframe:
  master_dist_MG[c("X1","X2","X3","X4","X5", "X6", "X7", "X8", "X9", "X10", "X11","X12","X13","X14","X15","X16","X17","X18","X19","X20", "X21","X22","X23","X24","X25","X26","X27","X28","X29","X30")]<-NULL
  ## For each Track2 find what is the minimal distance to MG and the max n of MG neiighbors in 30um diameter:
  master_distance_MG<-master_dist_MG%>%group_by(Track2, mouse, position, class)%>%summarize(n_MG=max(n_MG), min_MG =min(min_MG))
  ### plot per position types the values of 
  p <- ggplot(master_distance_MG)+geom_jitter(aes(x=class, y=n_MG))+
    geom_bar(aes(x=class , y=n_MG),alpha=0.5,stat = "summary") +ggtitle("number of neighboring MG cells per position type")+theme_classic()+
    theme(aspect.ratio=0.7)
  
  p

  p <- ggplot(master_distance_MG)+geom_jitter(aes(x=class, y=min_MG))+
    geom_boxplot(aes(x=class , y=min_MG),alpha=0.5, outlier.colour = NA) +ggtitle("min distance to MG cells per position type")+theme_classic()+
    theme(aspect.ratio=0.7)
  
  p

Time of interest selection and metadata info

Scroll for further detail

Time of interest selection and substitution of NA values:

### create empyt dataframe with combination of Time of interest:
  max(master_distance$Time)

## [1] 5.44

  length(seq(from=0, to=5.6, by=0.2))

## [1] 29

  empty_df = master_distance[1:29,] ### create a dataframe with the same size as the sequence of Time of interest:
  empty_df[!is.na(empty_df)] <- NA ## make it empyt
  empty_df$Time<-seq(from=0, to=5.6, by=0.2)
  empty_df$Track2<-0.5
  
  
  ### now merge this fake dataframe with my dataframe of interest:
  master_distance<-rbind(master_distance, empty_df)
  ### complete all the datasets so that they have the same timepoints:
  master_distance<-master_distance%>%complete(Track2,Time)
  ### remove the fake track that is not necessary anymore:
  
  master_distance<-subset(master_distance, !Track2==0.5)
  
  
  ##round the number, otherwise they are not equal
  master_distance$Time<-round(master_distance$Time,2)
  ##if there is any track2 and time that was duplicated remove the NA
  # Assuming your dataframe is named your_dataframe
  master_distance <- master_distance %>%
    group_by(Time, Track2) %>%
    dplyr::slice(if (n() == 1) 1 else which(!is.na(dist_tumor)))
  
  ### create a for loop to refill all the NA with interpolated values:
  ### create a list of the columns names that need to be refilled
  column_names<-names(master_distance)
  column_names<-subset(column_names,!column_names %in%c("TrackID","Track2", "ranks","ID","ID2","Time","position","mouse","class","pos", "Tracked","speed"))
  
  ## create a first dataset with refilled values for speed:
  time_series<-acast(master_distance, Time ~ Track2, value.var='speed',fun.aggregate = mean)
  ## rownames timepoints:
  row.names(time_series)<-unique(master_distance$Time)
  ## get rid of NA
  time_series_zoo<-zoo(time_series, row.names(time_series))
  time_series_zoo<-na.approx(time_series_zoo) ## replace by interpolated value
  time_series<-as.matrix(time_series_zoo)
  time_series2<-melt(time_series)
  data<-time_series2[complete.cases(time_series2), ] 
  colnames(data)<-c("Time", "Track2", "speed")
  ### ----------
  for (i in column_names){
    time_series<-acast(master_distance, Time ~ Track2, value.var=i,fun.aggregate = mean)
    row.names(time_series)<-unique(master_distance$Time)
    ## get rid of NA
    time_series_zoo<-zoo(time_series,row.names(time_series))
    time_series_zoo<-na.approx(time_series_zoo) ## replace by last value
    time_series<-as.matrix(time_series_zoo)
    time_series2<-melt(time_series)
    new<-time_series2[complete.cases(time_series2), ] 
    data[ , ncol(data) + 1] <- new[3]                  # Append new column
    colnames(data)[ncol(data)] <- paste0(i)
    
  }
  ### check that the data is correct by plotting the evolution of a certain variable overtime -----
  p1 <-ggplot(data, aes(Time, disp2, group=Track2, color = as.factor(Track2))) + 
    geom_smooth(method = "loess",size = 0.5, se = F, alpha=0.3, span=1)+
    theme_bw() +
    theme(axis.text.x = element_text(size=10), axis.text.y = element_text(size=5), axis.title.y = element_text(size = 10), axis.title.x = element_text(size = 15), legend.text=element_text(size= 10))+
    ggtitle("disp2")+theme(aspect.ratio=1,legend.position = "none")
  
  p1

Then, metadata information is added to the analysis:

### add metadata info
  
  master_cor <- data
  master_metadata<- master_distance[c("position","mouse","class", "Track2")]
  master_metadata<-na.omit(master_metadata)
  master_metadata<-master_metadata[!duplicated(master_metadata$Track2),]
 
  #detach(package:plyr, unload = TRUE)
  master_cor<- left_join(master_cor ,master_metadata)

  ### Filter timepoints of interest with same time interval:
  time_of_interest<-as.character(seq(from=0, to=5.6, by=0.2)) ### convert to character
  master_cor$Time<-as.character(master_cor$Time)
  master_cor<-subset(master_cor, Time%in%time_of_interest)
  master_cor$Time<-as.numeric(master_cor$Time) ### now back to numeric
  
  
  
  ##check that the imported tracks look ok
  g<-ggplot(master_cor)+geom_path(aes(x, y, group=Track2), size=0.2,
                                       arrow = arrow(ends = "last",type = "open",length = unit(0.01, "inches")))+
    theme_bw()+theme(aspect.ratio = 1)+facet_wrap(class~position,scales = "free")
  
  g

Finally, the generated dataframes are exported to use as Checkpoints when repeating the analysis:

Set working directory to export dataframes

setwd("D:/BEHAV3D_Tumor_Profiler/results/rds")

All imported information:

saveRDS(master_distance, "master_distance.rds")

Time of interest subset information:

saveRDS(master_cor,"master_cor.rds")

Microglia information:

saveRDS(master_distance_MG,"master_distance_MG.rds")

SR101 information:

saveRDS(master_distance_SR101,"master_distance_SR101.rds")

Blood Vessel information:

saveRDS(BV_df_sum,"BV_df_sum.rds")

2. You can continue with a previously started analysis using the .rds files obtained.

If the RDS files are in the working directory, the BEHAV3D Tumor Profiler script will skip the data input and initial pre-processing, by using the provided .rds files.

  setwd("D:/BEHAV3D_Tumor_Profiler/results")

  master_cor<-readRDS("master_cor.rds")
  master_distance<-readRDS("master_distance.rds")
  master_distance_MG<-readRDS("master_distance_MG.rds")
  master_distance_SR101<-readRDS("master_distance_SR101.rds")
  BV_df_sum<-readRDS("BV_df_sum.rds")
  print("data already imported")

## [1] "data already imported"

💻 Joint data processing

After the .rds checkpoints, we can continue with joint data processing.

In this section, the joint dataframes are adjusted so all track have the same length

### sumarize the length of the tracks:
## create relative time 
master_cor2<-master_cor %>% 
  group_by(Track2) %>%arrange(Time)%>%mutate(Time2 = Time - dplyr::first(Time))
ggplot(master_cor2,aes(Time2))+geom_histogram()+facet_wrap(.~ position)

This plot shows a summary of the imported tracks for each position and time

max_time<-master_cor2%>%group_by(position)%>%summarise(max_Time=max(Time2))
min_time <- min(max_time$max_Time) #this is the minimal time that any position has
hist(max_time$max_Time)

Here we can see the track length distribution, so we can normalize all tracks to the minimal common time:

master_cor2<-master_cor2 %>% 
  group_by(Track2)%>%filter(Time2<min_time)  ##Minimal time for a position is 2.6
## keep only the tracks that are of the same length  (max number of timepoints (13 in this case), this is how they were defined)
# Calculate the most frequent n for each group
subtrack_length<-master_cor2%>%group_by(Track2, position)%>%
  summarise(n=n_distinct(Time2), first_t=min(Time2), last_t=max(Time2))%>%
  ungroup()
hist(subtrack_length$n)

This histogram shows the track length across all tracks, so the most common track length can be selected

freq_n <- as.numeric(which.max(table(subtrack_length$n))) # Most frequent n
subtrack_length<-subtrack_length%>%filter(n==freq_n)  # we need to automate this to get the most frequent n
#Now we only keep these subtracks that have the same length:
master_cor2<-subset(master_cor2, Track2%in%subtrack_length$Track2)

Plot the unique tracks for each position using the tumor distance as color label:

### create a parameter for distance to tumor core
master_cor2<-master_cor2 %>% 
  group_by(position)%>%mutate(dist_tumor_1 = scales::rescale(dist_tumor, to=c(0,100)))

### normalize the distance to tumor factor per position position
master_cor2<-master_cor2 %>% 
  group_by(Track2) %>%arrange(Time2)%>%mutate(dist_tumor = dist_tumor - dplyr::first(dist_tumor))%>%ungroup()

#### check tracks
master_M4_test<-master_cor2%>%group_by(position, mouse)%>%mutate(x=x-min(x), y=y-min(y))
g<-ggplot(master_M4_test)+geom_path(aes(x, y, group=Track2, colour=dist_tumor), arrow = arrow(ends = "last", type = "closed",length = unit(0.005, "inches")))+theme_bw()
g<-g + theme(legend.position = "none")+scale_colour_gradient2(low = "blue", mid = "grey" , high = "red") +facet_wrap(.~position)
g

Number of unqiue tracks that are processed:

length(unique(master_cor2$Track2))

## [1] 981

Scaling

Now we need to do some feature engineering, since we make a sliding window analysis we need to requantify the values of displacement, speed, etc from scratch

## scaling
library(parallel)
library(dplyr)
library(dtwclust)
library(stats)
library(scales)
#detach(package:spatstat, unload = TRUE)
###normalize the data:
master_norm<-master_cor2%>%ungroup()
column_names2<-names(master_cor2)
## select what data to normalize
column_names2<-subset(column_names2,!column_names2 %in%c("Time","Track2","x","y","z","position","mouse","class","Time2", "dist_tumor_1", "dist_3_neigh", "dist_10_neigh", "iteration", "Track2"))
master_norm<-as.data.frame(master_norm)

# transform skwed variables
# Calculate skewness for each variable
skew_values <- apply(master_norm[c(column_names2)], 2, skewness)

# Identify variables with skewness greater than a threshold (e.g., 0.5)
skewed_variables <- names(skew_values[abs(skew_values) > 2])
# Apply logarithmic transformation to skewed variables
master_norm[ , skewed_variables] <- log1p(master_norm[ , skewed_variables])

PCA analysis

We performa a PCA dimensionality reduction to prepare for the multivariate analysis

### perfrom PCA on the factors I want to use
master_scaled<-master_norm[c(column_names2)]%>%mutate(across(everything(), scale))%>%ungroup() ##first scale
##now to give more weight to the varibale of dist_tumor so the ability of cells to leave
#master_scaled[,c("dist_tumor")]<-master_scaled[,c("dist_tumor")]*2

master_pca<-prcomp(master_norm[c(column_names2)], scale = T)

master_pca2<-master_pca[["x"]] ##select only PC
master_pca3<-data.frame(master_norm[,c(2)],master_pca2[,c(1:3)]) ### take only first 3 components since they explain most variation

colnames(master_pca3)[1]<- "Track2"

Plot the PCA information:

pcaCharts <- function(x) {
  x.var <- x$sdev ^ 2
  x.pvar <- x.var/sum(x.var)
  print("proportions of variance:")
  print(x.pvar)
  
  par(mfrow=c(2,2))
  plot(x.pvar,xlab="Principal component", ylab="Proportion of variance explained", ylim=c(0,1), type='b')
  plot(cumsum(x.pvar),xlab="Principal component", ylab="Cumulative Proportion of variance explained", ylim=c(0,1), type='b')
  screeplot(x, main="Absolute Variance", xlab="Principal Component")
  screeplot(x,type="l", main="Absolute Variance")
  par(mfrow=c(1,1))
}

pcaCharts(master_pca)

## [1] "proportions of variance:"
## [1] 0.5643365414 0.1929346874 0.1787666118 0.0631326010 0.0008295584

📊 Modules

Once we have normalized, adjusted and prepared all the data for analysis, it is time to proceed to the analysis modules.

The BEHAV3D Tumor Profiler is divided in three distinct modules:

1. Heterogeneity Module - Implements multiparametric single-cell time-series classification, allowing us to identify distinct single-cell behavioral patterns

Optional modules:

2. Large-scale phenotyping module - Performs large-scale TME phenotyping and identifies regions with a specific cellular composition and architecture within the TME of intravitally imaged tumors

3. Small-scale phenotyping module - Further refines TME phenotyping to better understand tumor cell behavior

The optional modules are already including the necessary sections for module 1 to allow for an easy run.

Modules are collapsed to facilitate navigation. Please unscroll the modules you would like to check

1. Heterogeneity Module

Click to expand the Heterogeneity module

Dynamic Time Warping

Dynamic Time Warping (DTW) is an algorithm used to measure similarity between two time series by aligning them in a way that minimizes the distance between corresponding points. It is particularly useful for comparing time series that may vary in speed or timing, like on this case.

If the matrix_distmat.rds file exists, it will skip the DTW analysis time, as it has been run previously. Else, it will run normally and generate the file at the end.

###MULTIVARIATE
list_master_pca <- split(master_pca3[,-c(1)],master_pca3$Track2) ## split into list

if ( ! file.exists("matrix_distmat.rds")){
  ##parallel working
  # load parallel
  library(parallel)
  # create multi-process workers
  workers <- makeCluster(detectCores()-2)
  # load dtwclust in each one, and make them use 1 thread per worker
  invisible(clusterEvalQ(workers, {
    library(dtwclust)
    RcppParallel::setThreadOptions(1L)
  }))
  # register your workers, e.g. with doParallel
  require(doParallel)
  registerDoParallel(workers)
  
  ###MULTIVARIATE
  set.seed(123)
  distmat <- proxy::dist(list_master_pca, method = "dtw")
  saveRDS(distmat, file = "distmat.rds")
  
  matrix_distmat<-as.matrix(distmat)
  saveRDS(matrix_distmat, file = "matrix_distmat.rds")
  

}else{
  matrix_distmat <- readRDS("matrix_distmat.rds")
  }


Track2<-as.numeric(names(list_master_pca))

UMAP and clustering

Now, we can perform the UMAP representation and cluster in the desired number of clusters. Any of the parameters can be modified to your convenience

# Adjust parameters according to the experiment characteristics
umap_dist<- umap(matrix_distmat,n_components=2,input="dist",init = "random", 
                 n_neighbors=7,min_dist=0.5, spread=6,n_epochs=1000 , local_connectivity=1) 

umap_1 <- as.data.frame(umap_dist$`layout`) 
#scatter3Drgl(umap_1$V1, umap_1$V2, umap_1$V3)

Plot of the raw UMAP

plot(umap_dist$`layout`, # Plot the result
     col=rgb(0,0,0,alpha=0.1), 
     pch=19,
     asp=0.4, xlab="UMAP 1", ylab="UMAP 2",
     main="Raw UMAP")

Here, we determine the number of clusters for the analysis. (n_cluster)

umap_1 <- as.data.frame(umap_dist$`layout`) 
Track2_umap<-cbind(Track2,umap_1)
positiontype<-master_norm[,c("Track2", "position","class","mouse")]
positiontype<- positiontype[!duplicated(positiontype$Track2),]
umap_2 <- left_join(Track2_umap ,positiontype)

# Clustering
n_cluster=7
hc <- hclust(dist(umap_dist$`layout`, method = "euclidean"), method = "ward.D2")
hierarchical_clusters <- cutree(hc, k = n_cluster) 
umap_3 <- cbind(hierarchical_clusters, umap_2)

colnames(umap_3)[1]<- "cluster2"

Here. we can obtain different UMAP representations, providing our color palette

##plot

# Make a color palette that adjusts to the cluster direction
# Generate a continuous palette with 100 colors
continuous_palette <- colorRampPalette(c("cyan4", "darkturquoise","darkorchid4","darkorchid1", "deeppink4","deeppink1", "goldenrod3", "gold"))(50)

# Select 8 colors from the continuous palette
mypalette_1<- continuous_palette[seq(1, length(continuous_palette), length.out = n_cluster)]

ggplot(umap_3, aes(x=V1, y=V2, color=as.factor(cluster2))) +  
  geom_point(size=2, alpha=0.8) + labs(color="cluster")+
  xlab("") + ylab("")+ 
  ggtitle("umap Cluster") +scale_color_manual(values=mypalette_1)+
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)

UMAP positions colored by large-scale regions

ggplot(umap_3, aes(x=V1, y=V2, color=as.factor(class))) +  
  geom_point(size=0.5, alpha=0.8) + labs(color="Environment_cluster")+
  xlab("") + ylab("")+ 
  ggtitle("umap Cluster") +scale_color_manual(values=c("cyan","gold","red"))+
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+facet_wrap(mouse~position)

UMAP colored by large-scale regions

ggplot(umap_3, aes(x=V1, y=V2, color=as.factor(class))) +  
  geom_point(size=2, alpha=0.6) + labs(color="Tumor region phenotype")+
  xlab("") + ylab("")+ 
  ggtitle("distribution of large scale regions")+scale_color_manual(values=c("cyan","gold","red"))+
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)

### UMAP direction

Here, we incorporate the MG and SR101 information to project the direction information into the UMAP

### calculate average stats per track
master_class <- left_join(master_cor2 ,umap_3[c(1:4)], by = c("Track2" = "Track2"))
master_class<-master_class %>% 
  filter(cluster2 != 0)
master_class$cluster2<-as.numeric(master_class$cluster2)
master_class_sum<-master_class%>%group_by(position,class,mouse, Track2)%>%
  arrange(Time)%>%summarize(cluster2=mean(cluster2), V1=mean(V1), V2=mean(V2),
                            dist_tumor=mean(dist_tumor), dist_3_neigh=mean(dist_3_neigh),
                            dist_10_neigh=mean(dist_10_neigh), speed=mean(speed), disp2=mean(disp2), 
                            disp_d=mean(disp_d),disp_l=mean(disp_l), movement=last(dist_tumor), 
                            distance_to_tumor=last(dist_tumor_1),raw_dist_tumor=last(dist_tumor), 
                            Time=first(Time))%>%ungroup()


## Join the information on the SR101 and MG
master_class_sum <- left_join(master_class_sum,master_distance_MG[c("Track2","n_MG","min_MG")] ,by = c("Track2" = "Track2"))
master_class_sum <- left_join(master_class_sum,master_distance_SR101[c("Track2","n_SR101","min_SR101")] ,by = c("Track2" = "Track2"))
master_class_sum <- left_join(master_class_sum,BV_df_sum ,by = c("Track2" = "Track2"))

mid <- 0
master_class_sum$movement2<-ifelse(master_class_sum$movement>0,"away from tumor", ifelse(master_class_sum$movement==0, "no movement", "towards the tumor"))
master_class_sum<-master_class_sum[!is.na(master_class_sum$cluster2),]

saveRDS(master_class_sum,"master_class_sum.rds")
write.csv(master_class_sum,"master_class_sum.csv")

And here is the resulting plot:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=movement)) +  
  geom_point(size=2, alpha=0.8) +labs(color="Direction")+
  xlab("") + ylab("") +
  ggtitle("Direction of movement relative to tumor core") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+theme(aspect.ratio=1)+
  scale_color_gradient2(midpoint = mid, low = "blue",mid="grey80",
                        high = "red3")

Heatmap

In this section, we project the relevant clustering features in a Heatmap

###Create heatmap
sum_all<-master_class_sum%>%group_by(cluster2)%>%
  summarise(displacement2 = median(disp2), 
            speed = median(speed),displacement_delta=median(disp_d),displacement_length=median(disp_l),max_speed=quantile(speed, 0.75), persistance=(displacement_length/displacement_delta),nearest_10_cell= median(dist_10_neigh),dist_tumor_core=median(distance_to_tumor),
            n_SR101=mean(n_SR101, na.rm = T),
            min_SR101=mean(min_SR101, na.rm = T),n_MG=mean(n_MG, na.rm = T),min_MG=mean(min_MG, na.rm = T),min_BV_dist=mean(BV_min, na.rm = T),sd_BV_dist=mean(BV_sd, na.rm = T),mean_BV_dist=mean(BV_mean, na.rm = T),
            movement=median(movement) )

Cluster_movement<-sum_all[,c(1,length(names(sum_all)))]
Cluster_movement <-Cluster_movement[order(Cluster_movement$movement),]

stdize = function(x, ...) {(x - min(x, ...)) / (max(x, ...) - min(x, ...))}
sum_all[-c(1)] <- lapply(sum_all[-c(1)], stdize, na.rm = T)



df<-sum_all[,-c(1)]
df<-df[order(match(rownames(df), Cluster_movement$cluster2)), , drop = FALSE]%>%ungroup()
df<-as.data.frame(df)
rownames(df)<-Cluster_movement$cluster2

We show all the features extracted in the analysis:

heat_m<-pheatmap(df, clustering_method = "complete", cluster_cols=F,cluster_rows=F,cellwidth=20, cellheight=20,
                 treeheight_row = 0, treeheight_col = 0,fontsize = 8,na_col = "grey50", main="Features",
                 angle_col = 90,color = colorRampPalette(c("deepskyblue1", "grey95", "deeppink2"))(50))

df_cl_features<-df[,c("displacement2",     "speed", "max_speed", "displacement_delta" ,  "displacement_length", "persistance", "movement")]
rownames(df_cl_features)<-rownames(df)

And a heatmap of the features used for clustering:

pheatmap(df_cl_features, clustering_method = "complete", cluster_cols=F,cluster_rows=F,cellwidth=20, cellheight=20,
                 treeheight_row = 0, treeheight_col = 0,fontsize = 8,na_col = "grey50", main="Features used for clustering",
                 angle_col = 90,color = colorRampPalette(c("deepskyblue1", "grey95", "deeppink2"))(50))

We also print tha Cluster movement data (towards or away from the tumor):

print(Cluster_movement)  #3 per cluster we can see the direction

## # A tibble: 7 × 2
##   cluster2 movement
##      <dbl>    <dbl>
## 1        6   -8.13 
## 2        5   -2.14 
## 3        1   -0.962
## 4        7   -0.537
## 5        3    0.995
## 6        4    1.61 
## 7        2    6.34

Movement within UMAP Clusters

Plots the clusters according to direction and other relevant features

## Movement within the clusters in UMAP

## plot clusters giving them a color according to direction
# Convert df to a data frame
# Add cluster2 as a column
df1<-as.data.frame(df)

df1 <- mutate(df1, cluster2 = row.names(df))

# Order df1 by movement and reorder cluster2 accordingly
df1 <- df1 %>%
  arrange(movement) %>%
  mutate(cluster2 = factor(cluster2, levels = unique(cluster2)))
##set first order clusters based on the heatmap:

UMAP representation with the behavioral clusters:

ggplot(master_class_sum, aes(x = V1, y = V2, color = as.factor(cluster2))) +  
  geom_point(size = 2, alpha = 1) + 
  labs(color = "Cluster") +
  xlab("") + 
  ylab("") +
  scale_color_manual(values = setNames(mypalette_1, levels(df1$cluster2))) +  # Match palette to cluster2 levels  ggtitle("UMAP Cluster") +
  theme_light(base_size = 20) +
  theme(axis.text.x = element_blank(),
        axis.text.y = element_blank(),
        aspect.ratio = 1)

UMAP representation with the behavioral clusters among mice:

ggplot(umap_3, aes(x=V1, y=V2, color=as.factor(cluster2))) +  
  geom_point(size=1, alpha=0.8) + labs(color="cluster")+
  xlab("") + ylab("")+ 
  ggtitle("umap Cluster among mice") +scale_color_manual(values = setNames(mypalette_1, levels(df1$cluster2))) +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+facet_wrap(mouse~.)

UMAP representation with the localization bistribution of each track:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=distance_to_tumor)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP localization distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")))

UMAP representation with the raw movement of each track:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=movement)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP raw_movement") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")))

UMAP representation with the speed distribution of each track:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=speed)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP speed distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")))

UMAP representation with the squared displacement disitribution of each track:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=disp2)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP disp2distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")))

UMAP representation with the displacement length of each track:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=disp_l)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP disp_length") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")))

Cluster distibution

Provides a pie chart of the cluster distribution along the mice experiments

### plot enrichment on edge and core:
Number_cell_exp<-umap_3%>%group_by(position,class,mouse)%>%
  summarise(total_cell = n())
Percentage_clus<-left_join(Number_cell_exp,umap_3)
Percentage_clus <- Percentage_clus%>%group_by(cluster2,position,class,mouse)%>%
  summarise(total_cell = mean(total_cell), num_cluster=n())%>%mutate(percentage=num_cluster*100/total_cell)%>%ungroup()
Percentage_clus_2 <- Percentage_clus%>%group_by(cluster2, position,class,mouse)%>%summarise(se_percentage=sd(percentage)/sqrt(length(percentage)),percentage = mean(percentage))
### Plot circular map
Percentage_clus_2$cluster2<-as.factor(Percentage_clus_2$cluster2)
Percentage_clus_2$cluster2<- factor(Percentage_clus_2$cluster2, levels = levels(df1$cluster2))

Behavioral cluster distribution in each mouse

Per2<-ggplot(Percentage_clus_2, aes(fill=as.factor(cluster2), y=percentage, x="")) + 
  geom_bar( stat="identity", position="fill", width = 0.5)+ coord_flip()+ scale_y_reverse()
Per2 <- Per2 + facet_grid(.~mouse)
Per2<-Per2+theme_void()+  scale_fill_manual(values=mypalette_1)+theme(strip.text.x = element_text(size = 8, angle = 90))
Per2

Mouse distribution in eacch behavioral cluster

Per4<-ggplot(Percentage_clus_2, aes(fill=as.factor(mouse), y=percentage, x="")) + 
  geom_bar( stat="identity", position="fill", width = 0.5)+ coord_flip()+ scale_y_reverse()
Per4 <- Per4 + facet_grid(cluster2~.)
Per4<-Per4+theme_void()+theme(strip.text.x = element_text(size = 8, angle = 90))
Per4

Differences based on clusters

This section analyzes the differences between clusters for different features.

Code section

### plot differences of the values per cluster
master_class_sum<-as.data.frame(master_class_sum)
master_class_sum$cluster2<-as.character(master_class_sum$cluster2)
df1$mean_movement<-df1$movement
master_class_sum2<-left_join(master_class_sum, df1[,c("cluster2","mean_movement")])
master_class_sum2$cluster2 = factor(master_class_sum2$cluster2, levels=df1$cluster2)


p_speed <- ggplot(master_class_sum2,aes(x=as.factor(cluster2) , y=speed, group=cluster2,fill=cluster2)) +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("speed per cluster")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))


p_direction <- ggplot(master_class_sum2,aes(x=as.factor(cluster2) , y=movement, group=cluster2,fill=cluster2)) +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("speed per cluster")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(-15,20))


p_dist3_neigth <- ggplot(master_class_sum2,aes(x=as.factor(cluster2) , y=dist_3_neigh, group=cluster2, fill=cluster2)) +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot(alpha=0.5, outlier.colour = NA)+ggtitle("dist 3 neigh per cluster")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(10,70))


p_dist_10_neigh <- ggplot(master_class_sum2,aes(x=as.factor(cluster2) , y=dist_10_neigh, group=cluster2,fill=cluster2)) +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("dist 10 neigh per cluster")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(20,90))


p_min_MG <- ggplot(subset(master_class_sum2, !is.na(min_MG)) ,aes(x=as.factor(cluster2) , y=min_MG, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("distance to closest MG")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(2,45))


p_min_SR101 <- ggplot(subset(master_class_sum2, !is.na(min_SR101)) ,aes(x=as.factor(cluster2) , y=min_SR101, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("distance to closest SR101")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,50))


p_n_MG <- ggplot(subset(master_class_sum2, !is.na(n_MG)) ,aes(x=as.factor(cluster2) , y=n_MG, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("n MG")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,15))



p_nSR101 <- ggplot(subset(master_class_sum2, !is.na(n_SR101)) ,aes(x=as.factor(cluster2) , y=n_SR101, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("n SR101")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))



p_BV_mean <- ggplot(subset(master_class_sum2, !is.na(BV_mean)) ,aes(x=as.factor(cluster2) , y=BV_mean, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("BV_mean")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))



p_BV_min <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(cluster2) , y=BV_min, group=cluster2, fill=cluster2))  +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("BV_min")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))



p_BV_contact <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(cluster2) , y=BV_contact, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("BV_contact")+theme_classic()+ scale_fill_manual(values=mypalette_1)+
  theme(aspect.ratio=0.7)


p_BV_sd <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(cluster2) , y=BV_sd, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("Standart deviation distance")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,1.7))

The output in this section shows a boxplot for each feature in each behavioral cluster

• Speed

p_speed

• Direction

p_direction

• Distance to 3 neighbors

p_dist3_neigth

• Distance to 10 neighbors

p_dist_10_neigh

• Minimal distance to an MG

p_min_MG

• Minimal distance to a SR101

p_min_SR101

• Number of MG

p_n_MG

• Number of SR101

p_nSR101

• Mean distance to BV

p_BV_mean

• Minimal distance to BV

p_BV_min

• BV contact

p_BV_contact

• BV standard deviation (sd)

p_BV_sd

Backprojection

Select a position of interest to save .txt information about the specified position of interest in a .txt file.
This code section uses the master_distance.rds file generated in the File Import section

# Determine position of interest
position_interest<-"2430F11.CL3_1"

names_cell<-master_class[!duplicated(master_class$mouse),]
names_cell$position

## [1] 2430F11.CL3_1 2430F13.CL2_2 2430F16.CL3_1 2430F17.CL3_2 2430F18.CL2_1
## [6] 2430F12.CL2_2
## 36 Levels: 2430F11.CL1_1 2430F12.CL1_1 2430F13.CL1_1 ... 2430F18.CL3_2

Tracks_1<-master_class[which(master_class$position==position_interest),]
Tracks_1<-Tracks_1[!duplicated(Tracks_1$Track2),c(2,which( colnames(master_class)=="cluster2" ))]
##Join with information on unique TrackID for that experiment

# Generated during the data import
master_distance<-readRDS("D:/BEHAV3D_Tumor_Profiler/results/master_distance.rds")
master_distance<-master_distance[c("TrackID", "Track2")]%>% distinct(TrackID, .keep_all = TRUE)
Tracks_1<-left_join(Tracks_1,master_distance)
Tracks_1$Track2<-NULL
Track_1_list<-split(Tracks_1,Tracks_1$cluster2)

write(paste(as.character(Track_1_list), sep="' '", collapse=", "), paste0(position_interest,".txt"))

Then, we plot the cell trajectories combined with the behavioral cluster information:

###Plot cell trajectories:

## bind cluster information data to original dataset:
Tracks_1<-master_class
###plot all the tracks
Tracks_1<-Tracks_1[!duplicated(Tracks_1$Track2),c("cluster2", "Track2")]

Tracks_1$cluster2<-as.factor(Tracks_1$cluster2)
master_traject<-left_join(master_cor2,Tracks_1)

master_traject<-na.omit(master_traject)
master_traject<-subset(master_traject, cluster2!=0)
master_traject<-master_traject%>%group_by(position, mouse)%>%mutate(x_new=x-min(x), y_new=y-min(y))

master_traject$cluster2 = factor(master_traject$cluster2, levels=df1$cluster2)

• Cell Trajectories in all positions in the dataset, colored by behavioral cluster:

g_all_track<-ggplot(master_traject)+geom_path(aes(x_new, y_new, group=Track2, 
                                        color=as.factor(cluster2)), size=0.2,
                                    arrow = arrow(ends = "last",type = "open",length = unit(0.01, "inches")))+scale_color_manual(values=mypalette_1 )+
  theme_bw()+theme(aspect.ratio = 1)+facet_wrap(class~position)
g_all_track

• Cell trajectories in the position of interest, colored by behavioral cluster:

g_track_int1<-ggplot(subset(master_traject, position%in%position_interest))+geom_path(aes(x_new, y_new, group=Track2, color=as.factor(cluster2)), size=0.5,arrow = arrow(ends = "last",type = "open",length = unit(0.00, "inches")))+
  scale_color_manual(values=mypalette_1)+ggtitle(paste0(position_interest))+
  theme_bw()+theme(aspect.ratio = 1)
g_track_int1

• Raw cell trajectories in the position of interest

g_track_int2<-ggplot(subset(master_traject, position%in%position_interest))+geom_path(aes(x_new, y_new, group=Track2), size=0.5,arrow = arrow(ends = "last",type = "open",length = unit(0.00, "inches")))+
  scale_color_manual(values=mypalette_1)+ggtitle(paste0(position_interest))+
  theme_bw()+theme(aspect.ratio = 1)
g_track_int2

---

1 + 2. Large-scale phenotyping Module

Click to expand the Large-scale phenotyping module

Dynamic Time Warping

Dynamic Time Warping (DTW) is an algorithm used to measure similarity between two time series by aligning them in a way that minimizes the distance between corresponding points. It is particularly useful for comparing time series that may vary in speed or timing, like on this case.

If the matrix_distmat.rds file exists, it will skip the DTW analysis time, as it has been run previously. Else, it will run normally and generate the file at the end.

###MULTIVARIATE
list_master_pca <- split(master_pca3[,-c(1)],master_pca3$Track2) ## split into list

if ( ! file.exists("matrix_distmat.rds")){
  ##parallel working
  # load parallel
  library(parallel)
  # create multi-process workers
  workers <- makeCluster(detectCores()-2)
  # load dtwclust in each one, and make them use 1 thread per worker
  invisible(clusterEvalQ(workers, {
    library(dtwclust)
    RcppParallel::setThreadOptions(1L)
  }))
  # register your workers, e.g. with doParallel
  require(doParallel)
  registerDoParallel(workers)
  
  ###MULTIVARIATE
  set.seed(123)
  distmat <- proxy::dist(list_master_pca, method = "dtw")
  saveRDS(distmat, file = "distmat.rds")
  
  matrix_distmat<-as.matrix(distmat)
  saveRDS(matrix_distmat, file = "matrix_distmat.rds")
  
  
}else{
  matrix_distmat <- readRDS("matrix_distmat.rds")
}


Track2<-as.numeric(names(list_master_pca))

UMAP and clustering

Now, we can perform the UMAP representation and cluster in the desired number of clusters. Any of the parameters can be modified to your convenience

# Modify the parameters according to the characteristics of the experiment
umap_dist<- umap(matrix_distmat,n_components=2,input="dist",init = "random", 
                 n_neighbors=7,min_dist=0.5, spread=6,n_epochs=1000 , local_connectivity=1)  

umap_1 <- as.data.frame(umap_dist$`layout`)

Here, we can see the Raw UMAP plot:

plot(umap_dist$`layout`, # Plot the result
     col=rgb(0,0,0,alpha=0.1), 
     pch=19,
     asp=0.4, xlab="UMAP 1", ylab="UMAP 2",
     main="Raw UMAP")

Here we determine the number of clusters for the analysis. (n_cluster)

umap_1 <- as.data.frame(umap_dist$`layout`) 

Track2_umap<-cbind(Track2,umap_1)
positiontype<-master_norm[,c("Track2", "position","class","mouse")]
positiontype<- positiontype[!duplicated(positiontype$Track2),]
umap_2 <- left_join(Track2_umap ,positiontype)

# Clustering
n_cluster=7
hc <- hclust(dist(umap_dist$`layout`, method = "euclidean"), method = "ward.D2")
hierarchical_clusters <- cutree(hc, k = n_cluster) 
umap_3 <- cbind(hierarchical_clusters, umap_2)

colnames(umap_3)[1]<- "cluster2"

UMAP positions colored by large-scale regions

ggplot(umap_3, aes(x=V1, y=V2, color=as.factor(class))) +  
  geom_point(size=0.5, alpha=0.8) + labs(color="Environment_cluster")+
  xlab("") + ylab("")+ 
  ggtitle("umap Cluster") +scale_color_manual(values=c("cyan","gold","red"))+
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+facet_wrap(mouse~position)

UMAP colored by large-scale regions

ggplot(umap_3, aes(x=V1, y=V2, color=as.factor(class))) +  
  geom_point(size=2, alpha=0.6) + labs(color="Tumor region phenotype")+
  xlab("") + ylab("")+ 
  ggtitle("distribution of large scale regions")+scale_color_manual(values=c("cyan","gold","red"))+
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)

UMAP direction

Here, we incorporate the MG and SR101 information to project the direction information into the UMAP

See deatils

### calculate average stats per track
master_class <- left_join(master_cor2 ,umap_3[c(1:4)], by = c("Track2" = "Track2"))
master_class<-master_class %>% 
  filter(cluster2 != 0)
master_class$cluster2<-as.numeric(master_class$cluster2)
master_class_sum<-master_class%>%group_by(position,class,mouse, Track2)%>%
  arrange(Time)%>%summarize(cluster2=mean(cluster2), V1=mean(V1), V2=mean(V2),
                            dist_tumor=mean(dist_tumor), dist_3_neigh=mean(dist_3_neigh),
                            dist_10_neigh=mean(dist_10_neigh), speed=mean(speed), disp2=mean(disp2), 
                            disp_d=mean(disp_d),disp_l=mean(disp_l), movement=last(dist_tumor), 
                            distance_to_tumor=last(dist_tumor_1),raw_dist_tumor=last(dist_tumor), 
                            Time=first(Time))%>%ungroup()


## Join the information on the SR101 and MG
master_class_sum <- left_join(master_class_sum,master_distance_MG[c("Track2","n_MG","min_MG")] ,by = c("Track2" = "Track2"))
master_class_sum <- left_join(master_class_sum,master_distance_SR101[c("Track2","n_SR101","min_SR101")] ,by = c("Track2" = "Track2"))
master_class_sum <- left_join(master_class_sum,BV_df_sum ,by = c("Track2" = "Track2"))

mid <- 0
master_class_sum$movement2<-ifelse(master_class_sum$movement>0,"away from tumor", ifelse(master_class_sum$movement==0, "no movement", "towards the tumor"))
master_class_sum<-master_class_sum[!is.na(master_class_sum$cluster2),]

saveRDS(master_class_sum,"master_class_sum.rds")
write.csv(master_class_sum,"master_class_sum.csv")

Cluster distibution

Provides a pie chart of the cluster distribution along the mice experiments

Show code

sum_all<-master_class_sum%>%group_by(cluster2)%>%
  summarise(displacement2 = median(disp2), 
            speed = median(speed),displacement_delta=median(disp_d),displacement_length=median(disp_l),max_speed=quantile(speed, 0.75), persistance=(displacement_length/displacement_delta),nearest_10_cell= median(dist_10_neigh),dist_tumor_core=median(distance_to_tumor),
            n_SR101=mean(n_SR101, na.rm = T),
            min_SR101=mean(min_SR101, na.rm = T),n_MG=mean(n_MG, na.rm = T),min_MG=mean(min_MG, na.rm = T),min_BV_dist=mean(BV_min, na.rm = T),sd_BV_dist=mean(BV_sd, na.rm = T),mean_BV_dist=mean(BV_mean, na.rm = T),
            movement=median(movement) )

Cluster_movement<-sum_all[,c(1,length(names(sum_all)))]
Cluster_movement <-Cluster_movement[order(Cluster_movement$movement),]

stdize = function(x, ...) {(x - min(x, ...)) / (max(x, ...) - min(x, ...))}
sum_all[-c(1)] <- lapply(sum_all[-c(1)], stdize, na.rm = T)



df<-sum_all[,-c(1)]
df<-df[order(match(rownames(df), Cluster_movement$cluster2)), , drop = FALSE]%>%ungroup()
df<-as.data.frame(df)
rownames(df)<-Cluster_movement$cluster2

## plot clusters giving them a color according to direction
# Convert df to a data frame
# Add cluster2 as a column
df1<-as.data.frame(df)

df1 <- mutate(df1, cluster2 = row.names(df))

# Order df1 by movement and reorder cluster2 accordingly
df1 <- df1 %>%
  arrange(movement) %>%
  mutate(cluster2 = factor(cluster2, levels = unique(cluster2)))

# Make a color palette that adjusts to the cluster direction
# Generate a continuous palette with 100 colors
continuous_palette <- colorRampPalette(c("cyan4", "darkturquoise","darkorchid4","darkorchid1", "deeppink4","deeppink1", "goldenrod3", "gold"))(50)

# Select 8 colors from the continuous palette
mypalette_1<- continuous_palette[seq(1, length(continuous_palette), length.out = n_cluster)]


### plot enrichment on edge and core:
Number_cell_exp<-umap_3%>%group_by(position,class,mouse)%>%
  summarise(total_cell = n())
Percentage_clus<-left_join(Number_cell_exp,umap_3)
Percentage_clus <- Percentage_clus%>%group_by(cluster2,position,class,mouse)%>%
  summarise(total_cell = mean(total_cell), num_cluster=n())%>%mutate(percentage=num_cluster*100/total_cell)%>%ungroup()
Percentage_clus_2 <- Percentage_clus%>%group_by(cluster2, position,class,mouse)%>%summarise(se_percentage=sd(percentage)/sqrt(length(percentage)),percentage = mean(percentage))

Plots the pie chart of:

• Behavioral cluster distribution in large-scale regions for every mouse experiment

### Plot circular map
Percentage_clus_2$cluster2<-as.factor(Percentage_clus_2$cluster2)
Percentage_clus_2$cluster2<- factor(Percentage_clus_2$cluster2, levels = levels(df1$cluster2))
Per1<-ggplot(Percentage_clus_2, aes(fill=as.factor(cluster2), y=percentage, x="")) + 
  geom_bar( stat="identity", position="fill", width = 0.5)+ coord_flip()+ scale_y_reverse()
Per1 <- Per1 + facet_grid(class~mouse)
Per1<-Per1+theme_void()+ scale_fill_manual(values=mypalette_1)+theme(strip.text.x = element_text(size = 8, angle = 90))
Per1

- Behavioral cluster distribution in large-scale regions

Per3<-ggplot(Percentage_clus_2, aes(fill=as.factor(cluster2), y=percentage, x="")) + 
  geom_bar( stat="identity", position="fill", width = 0.5)+ coord_flip()+ scale_y_reverse()
Per3 <- Per3 + facet_grid(class~.)
Per3<-Per3+theme_void()+theme(strip.text.x = element_text(size = 8, angle = 90))+ scale_fill_manual(values=mypalette_1)
Per3

Large-scale regions’ statistics

This section provides boxplot information about features in different large-scale regions

Show code

# Additional Statistics

# Create an empty dataframe to store results
results_df <- data.frame(cluster2 = character(), p_value = numeric(),contrast=character(), pairwise_p_values=numeric())

# Create an empty list to store plots
plots_list <- list()

# Iterate over unique clusters
for (cluster in unique(Percentage_clus_2$cluster2)) {
  # Subset data for the current cluster
  cluster_data <- subset(Percentage_clus_2, cluster2 == cluster)
  
  # Calculate mean percentage for each mouse and subtract from original percentage
  normalized_data <- cluster_data %>%
    group_by(mouse) %>%
    mutate(normalized_percentage = (percentage - mean(percentage)) / sd(percentage)) %>%
    ungroup()
  
  # Fit model with normalized percentage values
  model <- lm(normalized_percentage ~ class, data = normalized_data)
  
  # Extract estimates and p-value from model summary
  model_summary <- summary(model)
  p_value <- anova(model)$`Pr(>F)`[1]
  # Conduct pairwise comparisons using emmeans
  pairwise_comp <- emmeans(model, pairwise ~ class, adjust = "tukey")
  
  # Extract pairwise comparisons and p-values
  pairwise_p_values <- summary(pairwise_comp)[["contrasts"]][["p.value"]]
  contrast<-summary(pairwise_comp)[["contrasts"]][["contrast"]]
  # Append cluster name, estimates, and p-value to results dataframe
  results_df <- rbind(results_df, data.frame(cluster2 = cluster, p_value = p_value, contrast=contrast,pairwise_p_values=pairwise_p_values))
  # Create boxplot of normalized percentages by class
  plot <- ggplot(normalized_data, aes(x = class, y = normalized_percentage)) +
    geom_boxplot(width=0.5) +geom_jitter(width = 0.3)+
    labs(title = paste("Cluster:", cluster)) +
    theme_bw()+theme(aspect.ratio = 1)
  # Store plot in the list
  plots_list[[cluster]] <- plot
}

Display the results dataframe:

# Display the results dataframe
print(results_df)

##    cluster2   p_value  contrast pairwise_p_values
## 1         1 0.6467245 CL1 - CL2         0.7541313
## 2         1 0.6467245 CL1 - CL3         0.6534765
## 3         1 0.6467245 CL2 - CL3         0.9843553
## 4         2 0.1263177 CL1 - CL2         0.4110613
## 5         2 0.1263177 CL1 - CL3         0.1098208
## 6         2 0.1263177 CL2 - CL3         0.6749548
## 7         3 0.2518502 CL1 - CL2         0.8263790
## 8         3 0.2518502 CL1 - CL3         0.2398153
## 9         3 0.2518502 CL2 - CL3         0.4942696
## 10        4 0.2370106 CL1 - CL2         0.4103133
## 11        4 0.2370106 CL1 - CL3         0.9059262
## 12        4 0.2370106 CL2 - CL3         0.2350897
## 13        5 0.2568905 CL1 - CL2         0.3853240
## 14        5 0.2568905 CL1 - CL3         0.2589728
## 15        5 0.2568905 CL2 - CL3         0.9480454
## 16        6 0.2822381 CL1 - CL2         0.4004833
## 17        6 0.2822381 CL1 - CL3         0.9936312
## 18        6 0.2822381 CL2 - CL3         0.3152953
## 19        7 0.7861878 CL1 - CL2         0.9645045
## 20        7 0.7861878 CL1 - CL3         0.9092402
## 21        7 0.7861878 CL2 - CL3         0.7689239

Plot the statistics in large-scale regions (environmental clusters)

plots_list

## $`1`

## 
## $`2`

## 
## $`3`

## 
## $`4`

## 
## $`5`

## 
## $`6`

## 
## $`7`

Differences based on large-scale regions

This section analyzes the differences between large-scale regions for different features.

Code section

####Differences based on environmental clusters

### plot differences of the values per cluster
master_class_sum<-as.data.frame(master_class_sum)
master_class_sum$cluster2<-as.character(master_class_sum$cluster2)
df1$mean_movement<-df1$movement
master_class_sum2<-left_join(master_class_sum, df1[,c("cluster2","mean_movement")])
master_class_sum2$cluster2 = factor(master_class_sum2$cluster2, levels=df1$cluster2)


#### try an anova between clusters based on distance to speed
p_class_speed <- ggplot(master_class_sum2,aes(x=as.factor(class) , y=speed, group=class,fill=class)) +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))+ggtitle("speed per env cluster")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(0,20))


p_class_disp2 <- ggplot(master_class_sum2,aes(x=as.factor(class) , y=disp2, group=class,fill=class)) +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))+ggtitle("speed per env cluster")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(0,180))


p_class_movement <- ggplot(master_class_sum2,aes(x=as.factor(class) , y=movement, group=class,fill=class)) +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red")) +ggtitle("movement direction per env cluster")+theme_classic()+
  theme(aspect.ratio=1)+scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")))+coord_cartesian(ylim = c(-10,17))

The output in this section shows a boxplot for each feature in each behavioral cluster according to large-scale regions:

• Speed

p_class_speed

• Square displacement

p_class_disp2

• Movement

p_class_movement

---

1 + 3. Small-scale phenotyping Module

Click to expand the Small-scale phenotyping module

Dynamic Time Warping

Dynamic Time Warping (DTW) is an algorithm used to measure similarity between two time series by aligning them in a way that minimizes the distance between corresponding points. It is particularly useful for comparing time series that may vary in speed or timing, like on this case.

If the matrix_distmat.rds file exists, it will skip the DTW analysis time, as it has been run previously. Else, it will run normally and generate the file at the end.

###MULTIVARIATE
list_master_pca <- split(master_pca3[,-c(1)],master_pca3$Track2) ## split into list

if ( ! file.exists("matrix_distmat.rds")){
  ##parallel working
  # load parallel
  library(parallel)
  # create multi-process workers
  workers <- makeCluster(detectCores()-2)
  # load dtwclust in each one, and make them use 1 thread per worker
  invisible(clusterEvalQ(workers, {
    library(dtwclust)
    RcppParallel::setThreadOptions(1L)
  }))
  # register your workers, e.g. with doParallel
  require(doParallel)
  registerDoParallel(workers)
  
  ###MULTIVARIATE
  set.seed(123)
  distmat <- proxy::dist(list_master_pca, method = "dtw")
  saveRDS(distmat, file = "distmat.rds")
  
  matrix_distmat<-as.matrix(distmat)
  saveRDS(matrix_distmat, file = "matrix_distmat.rds")
  
  
}else{
  matrix_distmat <- readRDS("matrix_distmat.rds")
}


Track2<-as.numeric(names(list_master_pca))

UMAP and clustering

Now, we can perform the UMAP representation and cluster in the desired number of clusters. Any of the parameters can be modified to your convenience.

However, in this module, the output of the UMAP is not relevant, as it does not reflect any small-scale phenotype.

Show code

# Adjust parameters according to the characteristics of the experiment
umap_dist<- umap(matrix_distmat,n_components=2,input="dist",init = "random", 
                 n_neighbors=7,min_dist=0.5, spread=6,n_epochs=1000 , local_connectivity=1) 

umap_1 <- as.data.frame(umap_dist$`layout`) 

Track2_umap<-cbind(Track2,umap_1)
positiontype<-master_norm[,c("Track2", "position","class","mouse")]
positiontype<- positiontype[!duplicated(positiontype$Track2),]
umap_2 <- left_join(Track2_umap ,positiontype)

# Clustering
set.seed(123)
n_cluster=7
hc <- hclust(dist(umap_dist$`layout`, method = "euclidean"), method = "ward.D2")
hierarchical_clusters <- cutree(hc, k = n_cluster) 
umap_3 <- cbind(hierarchical_clusters, umap_2)
colnames(umap_3)[1]<- "cluster2"


### calculate average stats per track
master_class <- left_join(master_cor2 ,umap_3[c(1:4)], by = c("Track2" = "Track2"))
master_class<-master_class %>% 
  filter(cluster2 != 0)
master_class$cluster2<-as.numeric(master_class$cluster2)
master_class_sum<-master_class%>%group_by(position,class,mouse, Track2)%>%
  arrange(Time)%>%summarize(cluster2=mean(cluster2), V1=mean(V1), V2=mean(V2),
                            dist_tumor=mean(dist_tumor), dist_3_neigh=mean(dist_3_neigh),
                            dist_10_neigh=mean(dist_10_neigh), speed=mean(speed), disp2=mean(disp2), 
                            disp_d=mean(disp_d),disp_l=mean(disp_l), movement=last(dist_tumor), 
                            distance_to_tumor=last(dist_tumor_1),raw_dist_tumor=last(dist_tumor), 
                            Time=first(Time))%>%ungroup()


## Join the information on the SR101 and MG
master_class_sum <- left_join(master_class_sum,master_distance_MG[c("Track2","n_MG","min_MG")] ,by = c("Track2" = "Track2"))
master_class_sum <- left_join(master_class_sum,master_distance_SR101[c("Track2","n_SR101","min_SR101")] ,by = c("Track2" = "Track2"))
master_class_sum <- left_join(master_class_sum,BV_df_sum ,by = c("Track2" = "Track2"))

mid <- 0
master_class_sum$movement2<-ifelse(master_class_sum$movement>0,"away from tumor", ifelse(master_class_sum$movement==0, "no movement", "towards the tumor"))
master_class_sum<-master_class_sum[!is.na(master_class_sum$cluster2),]

saveRDS(master_class_sum,"master_class_sum.rds")
write.csv(master_class_sum,"master_class_sum.csv")

Heatmap

In this section, we project the relevant small-scale features in a Heatmap

###Create heatmap
sum_all<-master_class_sum%>%group_by(cluster2)%>%
  summarise(displacement2 = median(disp2), 
            speed = median(speed),displacement_delta=median(disp_d),displacement_length=median(disp_l),max_speed=quantile(speed, 0.75), persistance=(displacement_length/displacement_delta),nearest_10_cell= median(dist_10_neigh),dist_tumor_core=median(distance_to_tumor),
            n_SR101=mean(n_SR101, na.rm = T),
            min_SR101=mean(min_SR101, na.rm = T),n_MG=mean(n_MG, na.rm = T),min_MG=mean(min_MG, na.rm = T),min_BV_dist=mean(BV_min, na.rm = T),sd_BV_dist=mean(BV_sd, na.rm = T),mean_BV_dist=mean(BV_mean, na.rm = T),
            movement=median(movement) )

Cluster_movement<-sum_all[,c(1,length(names(sum_all)))]
Cluster_movement <-Cluster_movement[order(Cluster_movement$movement),]

stdize = function(x, ...) {(x - min(x, ...)) / (max(x, ...) - min(x, ...))}
sum_all[-c(1)] <- lapply(sum_all[-c(1)], stdize, na.rm = T)

df<-sum_all[,-c(1)]
df<-df[order(match(rownames(df), Cluster_movement$cluster2)), , drop = FALSE]%>%ungroup()
df<-as.data.frame(df)
rownames(df)<-Cluster_movement$cluster2

Heatmap of all the included features:

heat_m<-pheatmap(df, clustering_method = "complete", cluster_cols=F,cluster_rows=F,cellwidth=20, cellheight=20,
                 treeheight_row = 0, treeheight_col = 0,fontsize = 8,na_col = "grey50", main=" All features",
                 angle_col = 90,color = colorRampPalette(c("deepskyblue1", "grey95", "deeppink2"))(50))

df_other_features<-df[,!colnames(df)%in%c("displacement2", "speed", "max_speed", "displacement_delta" ,  "displacement_length", "persistance", "movement")]
rownames(df_other_features)<-rownames(df)

Heatmap of the small-scale features:

pheatmap(df_other_features, clustering_method = "complete", cluster_cols=F,cluster_rows=F,cellwidth=20, cellheight=20,
         treeheight_row = 0, treeheight_col = 0,fontsize = 8,na_col = "grey50",main="Small-scale TME features",
         angle_col = 90,color = colorRampPalette(c("deepskyblue1", "grey95", "deeppink2"))(50))

Movement within UMAP Clusters

Plots the behavioral clusters according to relevant small-scale features

## plot clusters giving them a color according to direction
# Convert df to a data frame
# Add cluster2 as a column
df1<-as.data.frame(df)

df1 <- mutate(df1, cluster2 = row.names(df))

# Order df1 by movement and reorder cluster2 accordingly
df1 <- df1 %>%
  arrange(movement) %>%
  mutate(cluster2 = factor(cluster2, levels = unique(cluster2)))
##set first order clusters based on the heatmap:

• UMAP representation of the number of SR101 in 30um radius distribution:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=n_SR101)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP number of SR101 in 30um radius distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")),na.value="NA")

• UMAP representation of the minimal distance to an SR101 in 30um radius distribution:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=min_SR101)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP min distance to SR101 in 30um radius distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = RColorBrewer::brewer.pal(4, "Spectral"),na.value="NA", trans="pseudo_log")

• UMAP representation of the minimal distance to an MG in 30um radius distribution:

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=min_MG)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP min distance to MG in 30um radius distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = RColorBrewer::brewer.pal(4, "Spectral"),na.value="NA", trans="pseudo_log")

• UMAP representation of the number of MG in 30um radius distribution

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=n_MG)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP number of MG in 30um radius distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")),na.value="NA")

• UMAP representation of the mean distance to BV (Blood Vessels)

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=BV_mean)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP mean distance to Blood vessels") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")),na.value="NA", trans="pseudo_log")

• UMAP representation of the mean contact with BV

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=BV_min)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP mean contact with Blood vessels") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")),na.value="NA")

• UMAP representation of the big neighbor distribution

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=dist_10_neigh)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP big neighbours distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")),trans="pseudo_log")

• UMAP representation of the close neighbor distribution

ggplot(master_class_sum, mapping=aes(x=V1, y=V2, color=dist_3_neigh)) +  
  geom_point(size=2, alpha=0.6) +
  xlab("") + ylab("") +
  ggtitle("UMAP close neighbours distribution") +
  theme_light(base_size=20) +
  theme(axis.text.x=element_blank(),
        axis.text.y=element_blank())+theme(aspect.ratio=1)+
  scale_color_gradientn(colours = rev(RColorBrewer::brewer.pal(4, "Spectral")),trans="pseudo_log")

Differences based on clusters

This section analyzes the differences between clusters for small-scale features.

Code section

### plot differences of the values per cluster
master_class_sum<-as.data.frame(master_class_sum)
master_class_sum$cluster2<-as.character(master_class_sum$cluster2)
df1$mean_movement<-df1$movement
master_class_sum2<-left_join(master_class_sum, df1[,c("cluster2","mean_movement")])
master_class_sum2$cluster2 = factor(master_class_sum2$cluster2, levels=df1$cluster2)


p_dist3_neigth <- ggplot(master_class_sum2,aes(x=as.factor(cluster2) , y=dist_3_neigh, group=cluster2, fill=cluster2)) +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot(alpha=0.5, outlier.colour = NA)+ggtitle("dist 3 neigh per cluster")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(10,70))


p_dist_10_neigh <- ggplot(master_class_sum2,aes(x=as.factor(cluster2) , y=dist_10_neigh, group=cluster2,fill=cluster2)) +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("dist 10 neigh per cluster")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(20,90))


p_min_MG <- ggplot(subset(master_class_sum2, !is.na(min_MG)) ,aes(x=as.factor(cluster2) , y=min_MG, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("distance to closest MG")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(2,45))


p_min_SR101 <- ggplot(subset(master_class_sum2, !is.na(min_SR101)) ,aes(x=as.factor(cluster2) , y=min_SR101, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("distance to closest SR101")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,50))


p_n_MG <- ggplot(subset(master_class_sum2, !is.na(n_MG)) ,aes(x=as.factor(cluster2) , y=n_MG, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("n MG")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,15))


p_nSR101 <- ggplot(subset(master_class_sum2, !is.na(n_SR101)) ,aes(x=as.factor(cluster2) , y=n_SR101, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("n SR101")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))


p_BV_mean <- ggplot(subset(master_class_sum2, !is.na(BV_mean)) ,aes(x=as.factor(cluster2) , y=BV_mean, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("BV_mean")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))


p_BV_min <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(cluster2) , y=BV_min, group=cluster2, fill=cluster2))  +geom_jitter(width=0.2, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("BV_min")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,20))


p_BV_contact <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(cluster2) , y=BV_contact, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot(outlier.colour = NA, alpha=0.5) +ggtitle("BV_contact")+theme_classic()+ scale_fill_manual(values=mypalette_1)+
  theme(aspect.ratio=0.7)



p_BV_sd <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(cluster2) , y=BV_sd, group=cluster2, fill=cluster2))  +geom_jitter(width=0.3, alpha=0.5) +
  geom_boxplot( outlier.colour = NA, alpha=0.5) +ggtitle("Standart deviation distance")+theme_classic()+
  theme(aspect.ratio=0.7)+ scale_fill_manual(values=mypalette_1)+coord_cartesian(ylim = c(0,1.7))

The output in this section shows a boxplot for each small-scale feature in each behavioral cluster

• Distance to 3 neighbors

p_dist3_neigth

• Distance to 10 neighbors

p_dist_10_neigh

• Minimal distance to an MG

p_min_MG

• Minimal distance to a SR101

p_min_SR101

• Number of MG

p_n_MG

• Number of SR101

p_nSR101

• Mean distance to BV

p_BV_mean

• Minimal distance to BV

p_BV_min

• BV contact

p_BV_contact

• BV standard deviation (sd)

p_BV_sd

Differences based on large-scale regions

This section analyzes the differences between large-scale regions for small-scale features.

Code section

####Differences based on environmental clusters

### plot differences of the values per cluster
master_class_sum<-as.data.frame(master_class_sum)
master_class_sum$cluster2<-as.character(master_class_sum$cluster2)
df1$mean_movement<-df1$movement
master_class_sum2<-left_join(master_class_sum, df1[,c("cluster2","mean_movement")])
master_class_sum2$cluster2 = factor(master_class_sum2$cluster2, levels=df1$cluster2)


p_class_dist_10neigh <- ggplot(master_class_sum2,aes(x=as.factor(class) , y=dist_10_neigh, fill=class)) +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))  +ggtitle("dist 10 neigh per cluster")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(20,70))


p_class_minMG <- ggplot(subset(master_class_sum2, !is.na(min_MG)) ,aes(x=as.factor(class) , y=min_MG, fill=class))  +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))  +ggtitle("distance to closest MG")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(0,50))


p_classmin_SR101 <- ggplot(subset(master_class_sum2, !is.na(min_SR101)) ,aes(x=as.factor(class) , y=min_SR101, fill=class))  +
  geom_violin(alpha=1, outlier.colour = NA) +stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))+ggtitle("distance to closest SR101")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(0,50))


p_class_n_MG <- ggplot(subset(master_class_sum2, !is.na(min_MG)),aes(x=as.factor(class) , y=n_MG, fill=class)) +
  geom_violin(alpha=1, outlier.colour = NA) +stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))+ggtitle("n of MG per cluster")+theme_classic()+
  theme(aspect.ratio=1)


p_classn_SR101 <- ggplot(subset(master_class_sum2, !is.na(min_SR101)),aes(x=as.factor(class) , y=n_SR101, fill=class)) +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+scale_fill_manual(values=c("cyan","gold","red"))+ggtitle("n SR101 per cluster")+theme_classic()+
  theme(aspect.ratio=1)


p_class_BV_min <- ggplot(subset(master_class_sum2, !is.na(BV_min)) ,aes(x=as.factor(class) , y=BV_min, fill=class))  +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+ scale_fill_manual(values=c("cyan","gold","red"))  +ggtitle("min distance to BV")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(0,18))


p_class_BV_mean <- ggplot(subset(master_class_sum2, !is.na(BV_mean)) ,aes(x=as.factor(class) , y=BV_mean, fill=class))  +
  geom_violin(alpha=1, outlier.colour = NA)+stat_summary(fun = median, geom = "crossbar",  size = 1, color = "grey25")+ scale_fill_manual(values=c("cyan","gold","red"))  +ggtitle("mean distance to BV")+theme_classic()+
  theme(aspect.ratio=1)+coord_cartesian(ylim = c(0,18))



pcor_MG_position <- ggplot(subset(master_class_sum2, !is.na(min_MG)),aes(x=distance_to_tumor , y=n_MG)) +geom_jitter()+
  geom_point(alpha=0.5,stat = "summary") +geom_smooth(method = "lm")+ggtitle("scatter plot movement vs min_MG")+theme_classic()+
  theme(aspect.ratio=1)

The output in this section shows a boxplot for each feature in each behavioral cluster according to large-scale regions:

• Distance to 10 neighbors

p_class_dist_10neigh

• Minimal distance to an MG

p_class_minMG

• Minimal distance to an SR101

p_classmin_SR101

• Number of MG

p_class_n_MG

• Number of SR101

p_classn_SR101

• Minimal distance to BV

p_class_BV_min

• Mean distance to BV

p_class_BV_mean

---

Session Information

## R version 4.4.3 (2025-02-28 ucrt)
## Platform: x86_64-w64-mingw32/x64
## Running under: Windows 11 x64 (build 26100)
## 
## Matrix products: default
## 
## 
## Random number generation:
##  RNG:     L'Ecuyer-CMRG 
##  Normal:  Inversion 
##  Sample:  Rejection 
##  
## locale:
## [1] LC_COLLATE=Spanish_Spain.utf8  LC_CTYPE=Spanish_Spain.utf8   
## [3] LC_MONETARY=Spanish_Spain.utf8 LC_NUMERIC=C                  
## [5] LC_TIME=Spanish_Spain.utf8    
## 
## time zone: Europe/Madrid
## tzcode source: internal
## 
## attached base packages:
## [1] parallel  stats     graphics  grDevices utils     datasets  methods  
## [8] base     
## 
## other attached packages:
##  [1] doParallel_1.0.17      iterators_1.0.14       foreach_1.5.2         
##  [4] zoo_1.8-13             viridis_0.6.5          viridisLite_0.4.2     
##  [7] umap_0.2.10.0          tidyr_1.3.1            spatstat_3.3-1        
## [10] spatstat.linnet_3.2-5  spatstat.model_3.3-4   rpart_4.1.24          
## [13] spatstat.explore_3.3-4 nlme_3.1-167           spatstat.random_3.3-2 
## [16] spatstat.geom_3.3-5    spatstat.univar_3.1-1  spatstat.data_3.1-4   
## [19] sp_2.2-0               scales_1.3.0           reshape2_1.4.4        
## [22] readr_2.1.5            RANN_2.6.2             plotly_4.10.4         
## [25] pheatmap_1.0.12        gtools_3.9.5           ggplot2_3.5.2         
## [28] emmeans_1.10.7         e1071_1.7-16           dtwclust_6.0.0        
## [31] dtw_1.23-1             proxy_0.4-27           dplyr_1.1.4           
## [34] DT_0.33               
## 
## loaded via a namespace (and not attached):
##  [1] deldir_2.0-4          gridExtra_2.3         rlang_1.1.6          
##  [4] magrittr_2.0.3        clue_0.3-66           flexclust_1.5.0      
##  [7] compiler_4.4.3        mgcv_1.9-1            png_0.1-8            
## [10] vctrs_0.6.5           stringr_1.5.1         crayon_1.5.3         
## [13] pkgconfig_2.0.3       fastmap_1.2.0         labeling_0.4.3       
## [16] promises_1.3.2        rmarkdown_2.29        tzdb_0.4.0           
## [19] bit_4.5.0.1           purrr_1.0.4           xfun_0.52            
## [22] modeltools_0.2-23     cachem_1.1.0          jsonlite_1.9.1       
## [25] goftest_1.2-3         later_1.4.1           spatstat.utils_3.1-2 
## [28] cluster_2.1.8         R6_2.6.1              bslib_0.9.0          
## [31] stringi_1.8.4         RColorBrewer_1.1-3    reticulate_1.41.0    
## [34] jquerylib_0.1.4       estimability_1.5.1    Rcpp_1.0.14          
## [37] knitr_1.49            tensor_1.5            splines_4.4.3        
## [40] httpuv_1.6.15         Matrix_1.7-2          tidyselect_1.2.1     
## [43] rstudioapi_0.17.1     abind_1.4-8           yaml_2.3.10          
## [46] codetools_0.2-20      lattice_0.22-6        tibble_3.2.1         
## [49] plyr_1.8.9            shiny_1.10.0          withr_3.0.2          
## [52] askpass_1.2.1         coda_0.19-4.1         evaluate_1.0.3       
## [55] RcppParallel_5.1.10   polyclip_1.10-7       pillar_1.10.1        
## [58] stats4_4.4.3          shinyjs_2.1.0         generics_0.1.3       
## [61] vroom_1.6.5           hms_1.1.3             munsell_0.5.1        
## [64] xtable_1.8-4          class_7.3-23          glue_1.8.0           
## [67] lazyeval_0.2.2        tools_4.4.3           data.table_1.17.0    
## [70] RSpectra_0.16-2       mvtnorm_1.3-3         grid_4.4.3           
## [73] crosstalk_1.2.1       colorspace_2.1-1      cli_3.6.4            
## [76] spatstat.sparse_3.1-0 gtable_0.3.6          sass_0.4.9           
## [79] digest_0.6.37         ggrepel_0.9.6         farver_2.1.2         
## [82] htmlwidgets_1.6.4     htmltools_0.5.8.1     lifecycle_1.0.4      
## [85] httr_1.4.7            mime_0.12             bit64_4.6.0-1        
## [88] openssl_2.3.2

Thank you for reading the BEHAV3D Tumor Profiler - Guided Tutorial.